Signal Detection Theory October 10, 2013 Some Psychometrics! Response data from a perception experiment is usually organized in the form of a confusion.

Slides:



Advertisements
Similar presentations
Signal Detection Theory. The classical psychophysicists believed in fixed thresholds Ideally, one would obtain a step-like change from no detection to.
Advertisements

10 / 31 Outline Perception workshop groups Signal detection theory Scheduling meetings.
© 2011 Pearson Education, Inc
McGraw-Hill Ryerson Copyright © 2011 McGraw-Hill Ryerson Limited. Adapted by Peter Au, George Brown College.
Statistical Significance What is Statistical Significance? What is Statistical Significance? How Do We Know Whether a Result is Statistically Significant?
PSYCHOPHYSICS What is Psychophysics? Classical Psychophysics Thresholds Signal Detection Theory Psychophysical Laws.
Sensation Perception = gathering information from the environment 2 stages: –Sensation = simple sensory experiences and translating physical energy from.
The standard error of the sample mean and confidence intervals
The standard error of the sample mean and confidence intervals
CS 8751 ML & KDDEvaluating Hypotheses1 Sample error, true error Confidence intervals for observed hypothesis error Estimators Binomial distribution, Normal.
Introduction to Biomedical Statistics. Signal Detection Theory What do we actually “detect” when we say we’ve detected something?
Statistical Significance What is Statistical Significance? How Do We Know Whether a Result is Statistically Significant? How Do We Know Whether a Result.
TECT: Kacelnik Package Individual and group decision making under risk. Are groups more or less efficient in handling risky decisions than individuals?
Intro to Statistics for the Behavioral Sciences PSYC 1900 Lecture 10: Hypothesis Tests for Two Means: Related & Independent Samples.
Evaluating Hypotheses
Psychophysical methods Lavanya Sharan January 26th, 2011.
Definitions Uniform Distribution is a probability distribution in which the continuous random variable values are spread evenly over the range of possibilities;
Chapter Sampling Distributions and Hypothesis Testing.
Independent Sample T-test Often used with experimental designs N subjects are randomly assigned to two groups (Control * Treatment). After treatment, the.
8-2 Basics of Hypothesis Testing
z-Scores What is a z-Score? How Are z-Scores Useful? Distributions of z-Scores Standard Normal Curve.
PSY 307 – Statistics for the Behavioral Sciences
Chapter 7 Probability and Samples: The Distribution of Sample Means
The standard error of the sample mean and confidence intervals How far is the average sample mean from the population mean? In what interval around mu.
Short Notes on Theory of Signal Detection Walter Schneider
Psychophysics 3 Research Methods Fall 2010 Tamás Bőhm.
Statistics 11 Hypothesis Testing Discover the relationships that exist between events/things Accomplished by: Asking questions Getting answers In accord.
© 2008 McGraw-Hill Higher Education The Statistical Imagination Chapter 9. Hypothesis Testing I: The Six Steps of Statistical Inference.
Think of a topic to study Review the previous literature and research Develop research questions and hypotheses Specify how to measure the variables in.
EDUC 200C Friday, October 26, Goals for today Homework Midterm exam Null Hypothesis Sampling distributions Hypothesis testing Mid-quarter evaluations.
Chapter 8 Introduction to Inference Target Goal: I can calculate the confidence interval for a population Estimating with Confidence 8.1a h.w: pg 481:
Thinking About Psychology: The Science of Mind and Behavior 2e Charles T. Blair-Broeker Randal M. Ernst.
Stats 95.
Chapter Normal Probability Distributions 1 of © 2012 Pearson Education, Inc. All rights reserved. Edited by Tonya Jagoe.
Slide 12.1 Judgment and Choice MathematicalMarketing Chapter 12 Judgment and Choice This chapter covers the mathematical models behind the way that consumer.
Copyright © 2010, 2007, 2004 Pearson Education, Inc. Chapter 6 Normal Probability Distributions 6-1 Review and Preview 6-2 The Standard Normal.
The Method of Constant Stimuli & Signal Detection Theory VISN2211 Sieu Khuu David Lewis.
Inference We want to know how often students in a medium-size college go to the mall in a given year. We interview an SRS of n = 10. If we interviewed.
Review of Chapters 1- 6 We review some important themes from the first 6 chapters 1.Introduction Statistics- Set of methods for collecting/analyzing data.
Signal detection theory Appendix Takashi Yamauchi Texas A&M University.
Research Design & Analysis 2: Class 23 Announcement re. Extra class: April 10th BAC 237 Discrete Trials Designs: Psychophysics & Signal Detection.
Signal Detection Theory I. Challenges in Measuring Perception II. Introduction to Signal Detection Theory III. Applications of Signal Detection Theory.
Fundamentals of Sensation and Perception EXPLORING PERCEPTION BY STUDYING BEHAVIOUR ERIK CHEVRIER SEPTEMBER 16 TH, 2015.
Statistical Inference Statistical Inference involves estimating a population parameter (mean) from a sample that is taken from the population. Inference.
Copyright © 2010, 2007, 2004 Pearson Education, Inc Section 8-2 Basics of Hypothesis Testing.
PPA 501 – Analytical Methods in Administration Lecture 6a – Normal Curve, Z- Scores, and Estimation.
Correlation Assume you have two measurements, x and y, on a set of objects, and would like to know if x and y are related. If they are directly related,
11/23/2015Slide 1 Using a combination of tables and plots from SPSS plus spreadsheets from Excel, we will show the linkage between correlation and linear.
Lecture 17 Dustin Lueker.  A way of statistically testing a hypothesis by comparing the data to values predicted by the hypothesis ◦ Data that fall far.
KNR 445 Statistics t-tests Slide 1 Introduction to Hypothesis Testing The z-test.
Nonparametric Tests of Significance Statistics for Political Science Levin and Fox Chapter Nine Part One.
Sensation Perception = gathering information from the environment 2 stages: –Sensation = simple sensory experiences and translating physical energy from.
Inferential Statistics Inferential statistics allow us to infer the characteristic(s) of a population from sample data Slightly different terms and symbols.
THE NORMAL DISTRIBUTION AND Z- SCORES Areas Under the Curve.
Hypothesis Testing Introduction to Statistics Chapter 8 Feb 24-26, 2009 Classes #12-13.
The Normal distribution and z-scores
Outline of Lecture I.Intro to Signal Detection Theory (words) II.Intro to Signal Detection Theory (pictures) III.Applications of Signal Detection Theory.
COURSE: JUST 3900 INTRODUCTORY STATISTICS FOR CRIMINAL JUSTICE Test Review: Ch. 4-6 Peer Tutor Slides Instructor: Mr. Ethan W. Cooper, Lead Tutor © 2013.
Intro to Statistics for the Behavioral Sciences PSYC 1900 Lecture 7: Regression.
GRAPPLING WITH DATA Variability in observations Sources of variability measurement error and reliability Visualizing the sample data Frequency distributions.
Hypothesis test flow chart
Signal Detection Theory March 25, 2010 Phonetics Fun, Ltd. Check it out:
Signal detection Psychophysics.
HYPOTHESIS TESTING FOR DIFFERENCES BETWEEN MEANS AND BETWEEN PROPORTIONS.
SIGNAL DETECTION THEORY  A situation is described in terms of two states of the world: a signal is present ("Signal") a signal is absent ("Noise")  You.
Signal Detection Theory October 5, 2011 Some Psychometrics! Response data from a perception experiment is usually organized in the form of a confusion.
Evaluating Hypotheses. Outline Empirically evaluating the accuracy of hypotheses is fundamental to machine learning – How well does this estimate its.
How do we make decisions about uncertain events?
Signal detection theory
Volume 16, Issue 20, Pages (October 2006)
Presentation transcript:

Signal Detection Theory October 10, 2013

Some Psychometrics! Response data from a perception experiment is usually organized in the form of a confusion matrix. Data from Peterson & Barney (1952) Each row corresponds to the stimulus category Each column corresponds to the response category

Detection In a detection task (as opposed to an identification task), listeners are asked to determine whether or not a signal was present in a stimulus. For example--do the following clips contain release bursts? Potential response categories: SignalResponse Hit: Present (in stimulus)“Present” Miss: Present“Absent” False Alarm:Absent“Present” Correct Rejection:Absent“Absent”

Confusion, Simplified For a detection task, the confusion matrix boils down to just two stimulus types and response options… (Response Options) PresentAbsent PresentHitMiss AbsentFalse AlarmCorrect Rejection (Stimulus Types) Notice that a bias towards “present” responses will increase totals of both hits and false alarms. Likewise, a bias towards “absent” responses will increase the number of both misses and correct rejections.

Canned Examples From the text: in session 1, listeners are rewarded for “hits”. The resultant confusion matrix looks like this: PresentAbsent Present8218 Absent4654 The “correct” responses (in bold) = = 136

Canned Examples In session 2, the listeners are rewarded for “correct rejections”… PresentAbsent Present5545 Absent1981 The “correct” responses (in bold) = = 136 Moral of the story: simply counting the number of “correct responses” does not satisfactorily tell you what the listener is doing… And response bias is not determined by what they can or cannot perceive in the signal.

Detection Theory Signal Detection Theory: a “parametric” model that predicts when and why listeners respond with each of the four different response types in a detection task. “Parametric” = response proportions are derived from underlying parameters Assumption #1: listeners base response decisions on the amount of evidence they perceive in the stimulus for the presence of a signal. Evidence = gradient variable. perceptual evidence

The Criterion Assumption #2: listeners respond positively when the amount of perceptual evidence exceeds some internal criterion measure. perceptual evidence criterion ( ) “present” responses “absent” responses evidence > criterion  “present” response evidence < criterion  “absent” response

The Distribution Assumption #3: the amount of perceived evidence for a particular stimulus includes random variation… and the variation is distributed normally. perceptual evidence FrequencyFrequency  The categorization of a particular stimulus will vary between trials.

Normal Facts The normal distribution is defined by two parameters: mean (= “average”) (  ) standard deviation (  ) The mean = center point of values in the distribution The standard deviation = “spread” of values around the mean in the distribution. standard deviation  standard deviation 

Comparisons Assumption #4: responses to both “absent” and “present” stimuli in a detection task will be distributed normally. Generally speaking: the mean of the “present” distribution will be higher on the evidence scale than that of the “absent” distribution. Assumption #5: both “absent” and “present” distributions will have the same standard deviation. (This is the simplest version of the model.)

Interpretation correct rejections false alarms misseshits criterion Important: the criterion level is the same for both types of stimuli… …but the means of the two distributions differ

Sensitivity The distance (on the perceptual evidence scale) between the means of the distributions reflects the listener’s sensitivity to the distinction. Q: How can we estimate this distance? A: We measure the distance of the criterion from each mean. We can use z-scores to standardize our distance measures! In normal distributions, this distance: determines the proportion of responses on either side of the criterion

Z-Scores Example 1: criterion at the mean  Z-score = 0 50% hits, 50% misses Hits Misses

Z-Scores Example 2: criterion one standard deviation below the mean  Z-score = % hits, 15.9% misses Hits Misses

Z-Scores Note: P(Hits) = 1-P(Misses)  z(P(Hits)) = z(1-P(Misses)) = -z(P(Misses)) In this case: z(84.1) = -z(15.9) = 1 Hits Misses

D-Prime D-prime (d’) is a measure of sensitivity. = perceptual distance between the means of the “present” and “absent” distributions. This perceptual distance is expressed in terms of z- scores. d’ ss nn

D-Prime d’ ss nn Hits d’ combines the z-score for the percentage of hits…

D-Prime z(P(H)) ss nn Hits d’ combines the z-score for the percentage of hits… with the z-score for the percentage of false alarms. False Alarms -z(P(FA)) d’ = z(P(H)) - z(P(FA))

D-Prime Examples 1.PresentAbsent Present8218 Absent4654 d’ = z(P(H)) - z(P(FA)) = z(.82) - z(.46) = (-.1) = PresentAbsent Present5545 Absent1981 d’ = z(P(H)) - z(P(FA)) = z(.55) - z(.19) = (-.878) = Note: there is no absolute meaning to the value of d-prime Also: NORMSINV() is the Excel function that converts percentages to z-scores. (qnorm() works in R)

Near Zero Correction Note: the z-score is undefined at 100% and 0%. Fix: replace perfect scores with a minimal deviation from the limit (.5% or 99.5%) PresentAbsent Present1000 Absent7228 d’ = z(P(H)) - z(P(FA)) = z(.995) - z(.72) = = 1.99

Near Zero Correction Also note that we do not normally deal with sets of responses that total to 100 in our experimental data! Here’s another example of the “fix” in which perfect scores are replaced with scores that are just half a response unit above or below the minimum and maximum scores, respectively. PresentAbsent Present200 Absent614 Replace 20 with 19.5, so P(H) = 19.5/20 =.975 d’ = z(P(H)) - z(P(FA)) = z(.975) - z(.3) = (-.52) = 2.48

Calculating Bias An unbiased criterion would fall halfway between the means of both distributions. No bias (λ u ): P (Hits) = P (Correct Rejections) Bias (λ b ): P (Hits) != P (Correct Rejections) u b

Calculating Bias Bias = distance (in z-scores) between the ideal criterion and the actual criterion Bias (  ) = -1/2 * (z(P(H)) + z(P(FA))) u b 

For Instance Let’s say: d’ = 2 An unbiased criterion would be one standard deviation from both means… z(P(H)) = 1z(P(FA)) = -1 z(P(H)) = 1  P(H) = 84.1% z(P(FA)) = -1  P(FA) = 15.9% Bias (  ) = -1/2 * (z(P(H)) + z(P(FA))) = -1/2 * (1 + (-1)) = -1/2 * (0) = 0

Wink Wink, Nudge Nudge Now let’s move the criterion over 1/2 a standard deviation… z(P(H)) = 1.5z(P(FA)) = -.5 z(P(H)) = 1.5  P(H) = 93.3% (cf. 84.1%) z(P(FA)) = -.5  P(FA) = 30.9%(cf. 15.9%) Bias (  ) = -1/2 * (z(P(H)) + z(P(FA))) = -1/2 * (1.5 + (-.5)) = -1/2 * (1) = -.5

Calculating Bias: Examples 1. PresentAbsent Present8218 Absent4654  = -1/2 * (z(P(H)) + z(P(FA)) = -1/2 * (z(.82) + z(.46)) = - 1/2 * ( (-.1)) = PresentAbsent Present5545 Absent1981  = -1/2 * (z(P(H)) + z(P(FA)) = -1/2 * (z(.55) + z(.19)) = - 1/2 * ( (-.878)) =.376 The higher the criterion is set, the more positive this number will be.

Peach Colo(u)rs Listeners could replay stimuli as many times as they liked. Order of pictures was counterbalanced across presentations.

Target identification significantly better than chance (p <.001) Difference in accuracy between IDS and ADS utterances was nearly signification (p =.056).

In terms of sensitivity (d’): Sensitivity significantly greater in IDS utterances! (p =.003)  The properties of Infant-directed speech provide cues to syntactic disambiguation.

In terms of bias (  ): IDS utterances induced a significantly greater bias towards NV responses (p =.032) Why? Perhaps duration differences between utterance types provide a clue…